Abstract
RATIONALE AND OBJECTIVES: Breast cancer remains a leading cause of cancer-related morbidity and mortality globally. This study aimed to develop and validate predictive models for ALN pCR following NAC in breast cancer patients. MATERIALS AND METHODS: We conducted a retrospective analysis involving 189 patients who were diagnosed with primary breast cancer at the Affiliated Hospital of Qingdao University. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) was utilized to assess the characteristics of the tumors. Tumor segmentation was performed using itk-SNAP software, followed by voxel clustering to identify distinct habitat-derived regions. Logistic regression (LR) and multilayer perceptron (MLP) models were constructed using these features. RESULTS: The classification model incorporating with habitat-based radiomic features demonstrating superior predictive performance (AUC of 0.88 in training and 0.81 in test for LR). A clinicopathologic signature that includes factors such as age, hormone receptor status, the Ki-67 index, and clinical stage was established, achieving in an AUC of 0.81. To construct a nomogram, we integrated habitat-derived radiomic signature with clinicopathologic signature. This nomogram attained an AUC of 0.92 for the training cohort and 0.89 for the test cohort. Furthermore, calibration and decision curve analyses confirmed the nomogram's reliability and practical applicability in clinical settings. CONCLUSION: In summary, our results indicate that radiomic features extracted from pre-NAC DCE-MRI can improve the predictive accuracy for ALN pCR following NAC in individuals diagnosed with breast cancer. This finding highlights the promise of personalized treatment strategies for individual patients.